Recently, a technology has been put into practical use, which detects abnormality before a system or device (hereinafter, called monitoring target) reaches its service stop by a failure during an operation of the monitoring target by the pattern recognition technology such as a Mahalanobis-Taguchi system as one of multivariate analysis techniques, and prompts an appropriate measure.
In this technology, after it is ensured that the monitoring target exhibits a normal behavior, the monitoring target is previously tested using various use cases, and a database (hereinafter, called reference database or reference data set) representing a normal space is established from the aggregate of multidimensional data (hereinafter, called operation data set) acquired by various sensors of the monitoring target. The normal space represented by the reference database and a currently acquired operation data set are analyzed by the pattern recognition technology described above. On the basis of the magnitude of a numerical value acquired as an analysis result, the numerical value indicating incompatibility between a reference value and the current monitoring target, abnormality is then detected.
When a reference database is established, every use case when the monitoring target is operated is assumed, and information from a sensor that is involved by the behavior of the monitoring target is acquired as an operation data set, to establish the reference database. In an actual operation of the monitoring target, however, an exceptional behavior of the monitoring target, which results from an event that could not been assumed when every use case was assumed, may occur due to an environment where the monitoring target is located, characteristics of a load connected to the monitoring target, and the like. When such an exceptional behavior increases the numerical value indicating incompatibility, there is a possibility that, though this behavior falls within the range of normality, this is erroneously detected to be abnormal.
In this regard, the technology of updating the reference database during the operation of the monitoring target has been developed. For example, Patent Document 1 discloses a technology in which a reference database is updated in an environment monitoring system that detects environmental abnormality by using the Mahalanobis-Taguchi system.
In this environment monitoring system, a Mahalanobis distance is calculated from a data set acquired in an actual monitoring, and whether the calculated Mahalanobis distance in the monitoring is close to a reference Mahalanobis distance or not is determined. When it is determined to be close, the data set providing the Mahalanobis distance in the monitoring is newly added to an existing reference database to update the reference database, and when the reference Mahalanobis distance is updated, environmental abnormality is detected using the updated reference Mahalanobis distance.
In other words, when the numerical value indicating incompatibility of the operation data set to the reference database, the operation data set being acquired when the operation is monitored, is equal to or smaller than a certain reference, such a numerical value is allowed, and the reference database is updated so as to include the operation data set therein. In the case where a similar event occurs again, this event is not detected to be abnormal.    Patent Document 1: Japanese Patent Application Laid-open No. Hei 10-124766